A Novel Multi-core System for Fault Tolerance and Security with Diskless Checkpointing Capability

Improving fault tolerance within the clusters has become vital because of the drastic decrease in the Mean Time Between Failures (MTBF) in complex clusters. Checkpointing is one of the robust ways to improve fault tolerance by rolling back from a saved state in the event of a failure in the cluster. Since, checkpointing primarily relies on storage devices for storing the states at regular intervals it will be difficult to accommodate several storage devices to support the ever increasing number of components in the cluster. This performance bottleneck is overcome by an independent checkpointing method called, Diskless Checkpointing. Ever since diskless checkpointing was launched commercially there were several research explorations that led to the division of diskless checkpointing into three namely, neighbor based, parity based and Reed-Solomon code based. All these algorithms were applied for recovering clusters from multiple processor failures. On the other hand, due to space and power constrains the multicore architecture have started to be used widely in the clusters. The future world of supercomputing envisions that most of the upcoming clusters will contain a majority of multicore processors or Chip MultiProcessors(CMP). This shows the necessity for a robust diskless checkpointing algorithm for recovering from failures in multicore architecture. Thus this paper aims at investigating the existing diskless checkpointing algorithms for multiple processor failures and devising a robust system for CMP.